native-funsc.sh

#!/bin/bash

for SIMD in mmx sse sse2 sse3 ssse3 sse4.1 sse4.2 avx fma avx2 avx512bw avx512f avx512vl; do
  echo ${SIMD}
  gcc -DSIMDE_NO_NATIVE -DSIMDE_ENABLE_NATIVE_ALIASES -I. -fdump-tree-original ./test/build/CMakeFiles/munit.dir/munit/munit.c.o test/x86/${SIMD}.c -lm -o ${SIMD}.o || true  # 2> /dev/null || true
  grep Function ${SIMD}.c.004t.original | awk '{ print $3}' >> simde.functions
  while read -r line ; do
    if [[ "${line}" == "simde"* ]] ; then
        if grep -q " ${line##simde}" simde/x86/${SIMD}.h ; then
          echo ${line##simde}
        fi
    fi
  done < simde.functions >> simde.native.functions
done

sort -u < simde.native.functions > simde.sorted.native.functions

echo Found $(wc -l < simde.sorted.native.functions) native functions implemented

skips="_mm256_extract_epi8 _mm256_extract_epi16 _mm256_extract_epi32 _mm256_extract_epi64 _mm256_insert_epi8 _mm256_insert_epi16 _mm256_insert_epi32 _mm256_insert_epi64"
# These need to be skipped, or the loops manually unrolled.

for skip in ${skips} ; do sed -i "/${skip}/d" simde.sorted.native.functions ; done

for SIMD in mmx sse sse2 sse3 ssse3 sse4.1 sse4.2 avx fma avx2 avx512bw avx512f avx512vl; do
  while read -r line ; do
    sed -i -E "s/([^_])simde${line}/\1${line}/g" test/x86/${SIMD}.c
  done < simde.sorted.native.functions
done

For the first part, is there a reason not to just add -fdump-tree-original to the CFLAGS, then just compile normally?

Instead of hardcoding the ISA extensions, maybe something like for header in simde/x86/*.h; SIMD="$(basename "$header" .h)"; ...; done?

For generating the list of candidate functions, you could get rid of awk with something like grep -hoP '(?<=;; Function )simde_[^ ]+' "${SIMD}.c.004t.original". That wolud also let you get rid of the outer if.

Maybe it would be better to parse the function list from the XML data backing the Intel Intrinsics Guide? If you want, I can generate a list of native functions that we could just keep in the repo somewhere.